 |
Center for Nonlinear Studies |
In surveillance, exploration, environmental monitoring, and search and rescue/destroy, one is typically faced with a situation where a very large-scale domain needs to be surveyed using limited sensory resources. Such limitations include limited sensory ranges and a limited number of sensors. In these situations, any available immobile sensors will not be capable of covering the entire domain. Hence, one needs to mount sensors on mobile platforms and develop navigation algorithms that seek to survey the entire domain over time. This is the search problem. In addition to the search problem, the mobile platform is also required to perform other tasks such as classification and tracking of found objects of interest, or the return to a base station for refueling, among others. These tasks, in the worst case scenario, are competing. In other words, they can not all be performed at the same time. This introduces the problem of task decision-making: Which task should the mobile platform execute at any given moment of time? In this work, I will assume that the only two tasks to be executed are search and classification, which are two competing tasks since a mobile platform can perform either the search task (which requires mobility) or the classification task (which requires constraining the motion of the platform to that of the object being classified), but not both at the same time. This is a very critical decision as choosing one option over the other may result in missing other, more important objects not yet found, or missing the opportunity to satisfactorily classify a found critical object. Building on previous deterministic-based work, in this talk I will present Bayesian-based search versus classification decision-making criteria that result in guaranteed detection and classification of all objects in the domain. I will first present the results for a single mobile sensor platform and will then generalize the result to the centralized and decentralized (with intermittent communications) cases for multiple cooperating mobile platforms. I will briefly present some more recent unpublished work on the use of cost-aware Bayesian sequential risk analysis to address the same problem. I will conclude by presenting some recent results on applying some of these results to future space-augmented space surveillance networks.
Host: Garrett Kenyon, gkenyon@lanl.gov